Well... the issue is more mathematical than not. Two analog signals get together and heterodyne. They add together. Since our brains are the "vocoders," we can kind of pick out things, or at least recognize that two people are talking at the same time and ask one to standby and the other to say again. If one signal is stronger, a "capture effect" happens and the stronger signal covers the weaker signal completely (one of the reasons aviation sticks with AM).
In digital, there is that summing of two signals, too. What happens, though, is that a machine is attempting to reconstruct the signal and not someone's brain. Machines don't have that flexibility, unless there's an algorithm coded to do so. So, when a stream of data (and it's not purely 1's and 0's, in P25 phase I its four positions relative to zero, with each position representing a 11, 10, 01, or 00 in the bitstream) is sent, the result is destructive. The machine doesn't read it and stays muted instead. There is capture in digital, but Jack is right about that effect decreasing as bandwidth decreases. You saw this in simulcast systems, where one particular manufacture widened the P25 phase 1 signal to use less simulcast sites for capture. That doesn't fit the mask for narrowbanding, though, and some digital systems had to severely reconfigure to meet the mandate.
Digital DOES have forward error correction, and, for the 9,600 bits per second that are conveyed, about half of them are for embedded signaling and forward error correction.
There are workarounds for this limitation. 1) Training. Digital radio is not like analog, and some agencies make the mistake of assuming it is. Not stomping on other transmissions is not always common sense (especially if that's how the analog equipment has been used), it's learned. 2) Programming. You can program a digital radio to inhibit from talking over another digital signal from the same system. If the other user has an immediate emergency, they can quickly re-press their PTT button (also programmable) and talk uninhibited. But that has to be understood and set up by the vendor and system manager, and trained-on by the users.
Noise cancelling (phase cancellation, like Bose does) mics are in development for several of the manufacturers. I've heard them working and the results are better than when everyone started using digital. The question should probably be: "Why was digital ever released into production without putting it through its paces with real end-users, instead of just testing in a laboratory and then declaring great success?"